Sap Se

Sap Ag

Florian Kerschbaum

Andreas Schaad

Walter Tighzert

Mathias Kohler

Martin Haerterich

Axel Schroepfer

Florian Hahn

Isabelle Hang

Patrick Grofig

Philip A Robinson

Anselme Kemgne Tueno

Jochen Haller

Daniel Bernau

Jonas Boehler

Benny Fuhry

Yuecel Karabulut

Luciana Moreira Sa De Souza

Alessandro Sorniotti

Leonardo Weiss F Chaves

Philip-William Grassal

Jonas Janneck

Julien Jean-Pierre Vayssiere

Joern Kussmaul

Timon Hackenjos

Andreas Fischer

Nicolas Loza

Marek Jawurek

Benjamin Weggenmann

Nina Oertel

Leonardo Weiss Ferreira Chaves

Orestis Terzidis

Yordan Boev

Rafael José Deitos

Benny Goerzig

Jonas Robl

Lina Hirschoff

Mubashir Mehmood Qureshi

Wei Xu

Josef Köeble

Jayanth Jain Hassan Ajith Kumar

Nikolas Kraetzschmar

Lina Stehle

Josef Koeble

Terzidis Orestis

In an example embodiment, a differentially private function is computed via secure computation. Secure computation allows multiple parties to compute a function without learning details about the data. The differentially private function is performed via probability distribution, which then permits computation of a result that is likely to be very close to the actual value without being so exact that it can be used to deduce the underlying data itself.

Secure multiparty differentially private median computation

A non-interactive protocol is provided for evaluating machine learning models such as decision trees. A client can delegate the evaluation of a machine learning model such as a decision tree to a server by sending an encrypted input and receiving only the encryption of the result. The inputs can be encoded as vector of integers using their binary representation. The server can then evaluate the machine learning model using a homomorphic arithmetic circuit. The homomorphic arithmetic circuit provides an implementation that requires fewer multiplications than a Boolean comparison circuit. Efficient data representations are then combined with different algorithmic optimizations to keep the computational overhead and the communication cost low. Related apparatus, systems, techniques and articles are also described.

Private decision tree evaluation using an arithmetic circuit

Aspects of the current subject matter are directed to secure group file sharing. An architecture for end-to-end encrypted, group-based file sharing using a trusted execution environment (TEE) is provided to protect confidentiality and integrity of data and management of files, enforce immediate permission and membership revocations, support deduplication, and mitigate rollback attacks.

Secure group file sharing

Machine learning model data privacy can be maintained by training a machine learning model forming part of a data science process using data anonymized using each of two or more differential privacy mechanisms. Thereafter, it is determined, for each of the two or more differential privacy mechanisms, a level of accuracy and a level precision when evaluating data with known classifications. Subsequently, using the respective determined levels of precision and accuracy, a mitigation efficiency ratio is determined for each of the two or more differential privacy mechanisms. The differential privacy mechanism having a highest mitigation efficiency ratio is then incorporated into the data science process. Related apparatus, systems, techniques and articles are also described.

Differential privacy to prevent machine learning model membership inference

Provided is a system and method for executing an encrypted software program within a host platform. The execution may be bifurcated among a trusted module and an untrusted area of the host platform. In one example, the method may include receiving bytecode and encrypted data of a software program, decrypting, via a secure memory area, the encrypted data into decrypted data, executing, via the secure memory area, instructions from the bytecode on the decrypted data to generate execution results, encrypting the generated execution results, and transmitting the encrypted execution results to a remote computing device.

Executing encrypted data using secure enclave

Provided is a system and method for generating a subset of optimal variations of a software program which allow some statements of the control flow to be exposed to side channels. Furthermore, the subset of optimal variations may be selected based on a security and a performance trade-off analysis. In one example, the method may include identifying a set of statements within a control flow of a software program, generating a plurality of variations of the software program which comprise different subsets of statements which are exposed to side channels, respectively, determining one or more pareto-optimal variations of the software program based on side channel leakage values and performance values of the plurality of variations of the software program, and outputting information about the one or more pareto-optimal variations of the software program to a user device.

Generation of optimal program variation

Methods, systems, and computer-readable storage media for range queries over encrypted data include actions of receiving a range query token, determining one or more of whether a tree list of an encrypted search index is empty and a range of the token intersects with a range accounted of a tree in the tree list, the encrypted search index including the tree list and a point list, receiving encrypted query results based on one of a search tree, if the tree list is not empty and a range of the token is at least a sub-range of a range accounted for in the tree list, and the point list, if the tree list is empty or the range of the token is not at least a sub-range of a range accounted for in the tree list, and updating the encrypted search index based on the token.

Poly-logarithmic range queries on encrypted data

A garbled circuit and two garbled inputs are received by a server from each pair of a plurality of clients. The garbled circuit encodes a comparison function and the garbled inputs encode a respective data value from each of the clients in each pair. Thereafter, the server evaluates the garbled circuits using the corresponding garbled inputs to result in a plurality of comparison bits. The server can then sort the datasets in an ascending or descending order by using the comparison bits to compute the rank of each data value. Using the sorted datasets, the server determines a median value for the datasets and transmits data characterizing the median value to each of the clients.

Cloud-based secure computation of the median

Techniques for efficient, accurate, and secure computation of a differentially private median of the union of two large confidential datasets are disclosed. In some example embodiments, a computer-implemented method comprises obtaining secret shares of a first dataset of a first entity, secret shares of a second dataset of a second entity, secret shares of gap values for the first dataset, secret shares of gap values for the second dataset, secret shares of probability mass values for the first dataset, and secret shares of probability mass values for the second dataset. The probability mass values may be computed via an exponential mechanism. In some example embodiments, the computer-implemented method further comprises determining a median of a union of the first dataset and the second dataset using an inverse transform sampling algorithm based on the obtained secret shares, and then performing a function of a networked computer system using the determined median.

Secure sublinear time differentially private median computation

A server receives a corresponding data value encrypted using a common threshold public key from each of a plurality of clients. The server distributes the received data values to the clients for evaluating comparison of values. The server receives the encrypted comparison results from each of the clients in response to the distribution of the received encrypted data values. The comparison results are encrypted using the common key. The server homomorphically determines a ciphertext encrypting the rank of each client's data value using the comparison results. Further, the server can compute a ciphertext encrypting the median of the datasets. Thereafter, the server can initiate a threshold decryption to generate a final result.

Efficient cloud-based secure computation of the median using homomorphic encryption

Embodiments offer database security utilizing dictionary encoding, with certain functionality being implemented inside a secure environment, e.g., a Trusted Execution Environment (TEE). In particular, the secure environment receives a secret key from a data owner, and receives an encrypted query range and a dictionary reference from a query engine. Based upon the query range decrypted using the secret key, and also the dictionary loaded from a database, the secure environment searches the dictionary to produce list of value identifiers corresponding to the query range. The value identifiers are communicated outside the secure environment to the query engine for further processing (e.g., to generate RecordIDs), ultimately producing a query result for a user. Particular embodiments may leverage the processing power of an in-memory database engine in order to perform the role of the query engine that interacts with the secure environment.

Secure database utilizing dictionary encoding

A method of producing an anonymized vector for a text mining task in lieu of a feature vector is disclosed. A vocabulary is created from a corpus of documents, each of the corpus of documents having a context that is similar to a set of target documents. The set of target documents is received. The feature vector is generated from a first document of the set of target documents. The feature vector is transformed into a composition vector. A synthetic vector is constructed based on the composition vector. The synthetic vector is shared as the anonymized vector in lieu of the feature vector.

Method and system for automated text anonymization

Secure substring searching on encrypted data may involve a first preprocessing comprising fragmenting a plaintext string slated for remote secure storage, in a plurality of overlapping plaintext substrings. A second preprocessing encrypts these substrings into ciphertexts (e.g., utilizing Frequency-Hiding Order Preserving Encryption) further including position information of the substring. A search index and a secret state result from the first and second preprocessing. The ciphertexts and search index are outsourced to a database within an unsecure server. An engine within the server determines candidate ciphertexts matching a query request received from a secure client. The engine returns ciphertexts to the client for decryption according to the secret state. Preprocessing may be delegated to a third party for outsourcing search index/ciphertexts to the server, and the secret state to the client. Filtering of candidate ciphertexts on the server-side, can eliminate false positives and reduce the volume of communication with remote clients.

Secure substring search to filter encrypted data

Disclosed herein are system, method, and computer program product embodiments for secure data aggregation in databases. An embodiment operates by identifying a value column and a group column of a plurality of columns of a dataset. Two distinct group values of the group column are identified. A first group value is replaced with a first substitute value, and a second group value is replaced with a second substitute value. A value of the value column of each of the plurality of records and the substitute values are encrypted. The plurality of encrypted records are uploaded to a server.

Secure data aggregation in databases using dynamic bucketization and grouping bucketization

Embodiments implement leakage-free order-preserving encryption by assigning a distinct ciphertext for each plaintext, including repeated plaintext whose ciphertext is randomly inserted. In order to conceal insertion order, the randomized ciphertexts are compressed to minimal ciphertext space. A uniform distribution is achieved by rotating about a modulus on the ciphertexts rather than the plaintexts. The resulting ciphertext distribution has no leakage from the ciphertexts&#x2014;even if an adversary has perfect background knowledge on the distribution of plaintexts. The encryption may be further secured even against passive query monitoring attacks by hiding the access pattern using &#x3f5;, &#x3b4;-differential privacy, such that the adversary observing a sequence of queries will not learn the frequency of plaintext. The leakage-free order-preserving encryption may be converted into an adjustable encryption scheme to allow querying (e.g., on a remote server).

Leakage-free order-preserving encryption

Growing community of inventors

Karlsruhe, Germany

Florian Kerschbaum